17 public posts in this discussion.
Aussie Aussie Aussie! Damn we rock!
Australian Centre for Visual Technologies | VideoTrace: Rapid interactive scene modelling from video
These nutbars have created a tool (application, not a Korny) that lets you convert objects from 2d video footage into a 3d model
that is ****ing unreal I like! I like alot!
That's pretty cool. Great stuff!!
thats just awesome... you could seriously add yourself to anywhere at anytime..
just a comp and a camera...
Wow, I'm impressed.
That looks really quite easy to use too.
That is ****ing impressive, totaly amazing
hah - good thing you have that new sig newbit... i always thought your nick was: Newb IT Lord =)
That's amazing!
Great program. Well done to all at University of Adelaide.
The idea has been around for years but using just photos: PhotoModeler - close-range photogrammetry software
yeah, and Canoma has done it with photos for years too... but this is moving footage - you might as well compare my mini radio control car to a Koenigsegg CCX .. no comparison!
Thats impressive. Beats the SHIT out of traditional modelling - how long would those models have taken by hand?
Keep in mind that the models can only be as good as the source file. In other words there would be no provision to make the model bigger than it already is within the video as you'd lose quality. It's still not as powerful as modelling from scratch, but certainly more efficient for a lot of applications. If you had a video camera that could film say 4 times the quality of full HD and a good filming environment (say a studio) then it would certainly be fantastic and much more cost effective than traditional 3D modelling.
True the source is a video, however the concept is still the same, it is being modelled from a still image. The only difference I see is that it is done on the fly, but again that is nothing new, as it is the same process used to extract geographic data from aerial images.
actually it's not being modeled from a still image - it's being modeled from paused video - yeah sounds the same i know, but the difference is the motion tracking features which move the object's control points on the fly as you skip through the frames, automatically adjusting sizing info. I refer you again to my analogy of the RC car vs supercar. =)
As for geographic stuffs ---> 3d, isn't that an automated process that just converts height information in the form of gradient maps onto some sort of deformation mesh? which, if I'm right, is totally different and something you could fudge with good old Bryce 3d without breaking a sweat :)
You are quite correct Digital Terrain Models are gradient maps, however they are produced in the same way, using 2 images (usually satellite) of the same area at different angles.
It is called photogrametry, stereophotogramtry is still used quite a lot to create topographic data. You may have seen the Cyclone Tracy display at the museum? They used to have two aerial photos with stero glasses so that when you look at the image it appears 3D.
I'm not knocking what they have done I am just saying that it is not a new concept and there have been photogrammetry workstations around for a while that do a similar thing.
The big difference here is its calculated and recalculated on the fly with the motion, not just static images with the end result being a full 3D model from all angles, not a single "looks 3D" stereoscopic picture.